Tunning Pss Using GA
Transcript of Tunning Pss Using GA
-
8/9/2019 Tunning Pss Using GA
1/7
ELSEVIER
ELECTRIC
POUIER
SY rems
RESEIMH
Electric Power System s Research 39 (1996) 137 143
Tuning of power system stabilizers using genetic algorithms
Y.L. Abdel-Magid *, M.M. Dawoud
Electrica l Engineering Department. King Fahd Lnicersit~~ c~f Petroleum und Minerals, Dhahran 31261, Saudi Arabia
Received 21 June 1996: accepted 3 July 1996
Abstract
Several techniques exist for developing optimal controllers. This paper investigates the tuning of power system stabilizers (PSS)
using genetic algorithms (GA). A digital simulation of a linearized model of a single-machine infinite bus power system at some
operating point is used in conjunction with the genetic algorithm optimization process. The integral of the square of the error and
the time-multiplied absolute value of the error performance indices are considered in the search for the optimal PSS parameters.
In order to have good damping characteristics over a wide range of operating conditions, the PSS parameters are optimized
off-l ine for a selected set of grid points in the real power (P)-reactive power (Q) domain. The optimal settings thus obtained can
then be stored and retrieved on-line to update the PSS parameters based on measurements of the generator real and reactive
power. T ime domain simulations of the system with GA-tuned PSS show the improved dynamic performance under widely
varying load conditions. 0 1996 Elsevier Science S.A.
Keywords: Dynamic stability; Genetic algorithm: Power system stabilizer
1. Introduction
The application of genetic algorithms (GA) has re-
cently attracted the attention of researchers in the con-
trol area [l-4]. From the literature it is clearly seen that
genetic algorithms can provide powerful tools for opti-
mization. The present paper demonstrates the use of
GA to tune the parameters of a power system stabilizer
(PSS).
The use of high-speed excitation systems has long
been recognized as an effective method of increasing
stability limits. Static excita tion systems appear to offer
the practical ultimate in high-speed performance,
thereby providing a gain in stability limits. Unfortu-
nately. the high speeds and gains that give them this
capability also result in poor system damping under
certain conditions of loading [5]. To offset this effect
and to improve the system damping in general, supple-
mentary stabilizing signals are introduced in the excita-
tion systems through fixed parameters lead/lag power
system stabilizers [6,7]. The parameters of the PSS are
normally fixed at certain values which are determined
under a particular operating condition.
* Corresponding author. Tel.: + 966 3 8602277: fax: + 966 3
8603535.
037%7796:96 15.00 8 1996 Elsevler Scienc e S..4 Al l rights reserved.
PI1 SO378-7796(96)01 105-4
It is important to recognize that machine parameters
change with loading, making the dynamic behavior of
the machine quite different at different operating
points. Since these parameters change in a rather com-
plex manner [8,9], a set of PSS parameters which
provide good dynamic performance under a certain
operating condition may no longer yield satisfactory
results when there is a drastic change in the operating
point. Therefore, it is necessary to adapt the PSS
parameters in real time based on measurements of the
machine loading. In this paper, the optimum values of
the PSS parameters are determined off-line using a
genetic algorithm. The procedure is repeated for a
selected set of grid points in the real power/reactive
power domain. The optimum settings thus obtained can
be stored in a look-up table . Based on the on-line
measurement of values of P and Q, the PSS parameters
are updated in step with changes in the operating
conditions in the ensuing steady state [lo].
Scalar integral performance indices have proved to
be the most meaningful and convenient measures of
dynamic performance [ 11,121. Two performance indices
have been chosen for this study: they are the popular
integral of the squared error (BE) and the integral of
time-multiplied absolute value of the error (ITAE), and
are given, respectively, by
-
8/9/2019 Tunning Pss Using GA
2/7
138
Y.L. Abdel-Magid, M.M. Dawoud : Electric Power Systems Research 39 (1996) 137- 143
s,= x
s
e(t) dr
(1)
0
s, =
s
x
+O)l dt (2)
0
The system to be studied is that of a single machine
connected to an infinite bus through a transmission
line. The PSS considered is a derivative-type power
stabilizer.
A digital simulation is used in conjunction with the
genetic algorithms optimization process to determine
the optimum parameters of the PSS for the perfor-
mance indices considered. Genetic algorithms are used
as parameter search techniques which utilize the genetic
operators to find near-optimal solutions. The advantage
of the GA technique is that it is independent of the
complexity of the performance index considered. It
suffices to specify the objective function and to place
finite bounds on the optimized parameters.
2. Genetic algorithms
Genetic algorithms (GA) are global search tech-
niques, based on the operations observed in natural
selection and genetics [l]. They operate on a population
of current approximations-the individuals-initially
drawn at random, from which improvement is sought.
Individuals are encoded as strings (chromosomes) con-
structed over some particular alphabet, e.g., the binary
alphabet {0, 1 , so that chromosome values are
uniquely mapped onto the decision variable domain.
Once the decision variable domain representation of the
current population is calculated, individual perfor-
mance is assumed according to the objective function
which characterizes the problem to be solved. It is also
possible to use the variable parameters directly to repre-
sent the chromosomes in the GA solution.
At the reproduction stage, a fitness value is derived
from the raw individual performance measure given by
the objective function, and used to bias the selection
process. Highly fit individuals will have increasing op-
portunities to pass on genetically important material to
successive generations. In this way, the genetic al-
gorithms search from many points in the search space
at once and yet continually narrow the focus of the
search to the areas of the observed best performance.
The selected individuals are then modified through
the application of genetic operators, in order to obtain
the next generation. Genetic operators manipulate the
characters (genes) that constitute the chromosomes di-
rectly, following the assumption that certain genes
code, on average, for fitter individuals than other genes.
Genetic operators can be divided into three main cate-
gories [2]: selection, crossover and mutation.
infinite bus
Fig. 1. Single machine connected to an infinite bus.
1. Selection: selects the fittest individuals in the current
population to be used in generating the next popula-
tion.
2. Cross-over: causes pairs, or larger groups of individ-
uals, to exchange genetic information with one an-
other.
3. Mutation: causes individual genetic representations
to be changed according to some probabilistic rule.
Genetic algorithms are more likely to converge to
global optima than conventional optimization tech-
niques, since they search from a population of points,
and are based on probabilistic transition rules. Conven-
tional optimization techniques are ordinarily based on
deterministic hill-climbing methods, which, by defini-
tion, will only find local optima. Genetic algorithms can
also tolerate discontinuities and noisy function evalua-
tions.
3. Problem formulation
The system considered in this paper is a synchronous
machine connected to an infinite bus through a trans-
mission line, as shown in Fig. 1. The synchronous
machine is described by a fourth-order model. The
relations in the block diagram shown in Fig. 2 apply to
a two-axis machine representation with a field circuit in
the direct axis but without damper windings. The inter-
action between the speed and voltage control equations
of the machine is expressed in terms of six constants
K,
-Kh.
These constants with the exception of
K3,
which
is only a function of the ratio of the impedance, are
dependent upon the actual real and reactive power
loading as well as the excitation levels in the machine
I I
Fig. 2. System block diagram
-
8/9/2019 Tunning Pss Using GA
3/7
Y.L. Abdel-Magid, M.M. Dawoud: Electru Power Systems Research 39 (1996) 137- 143
139
RANDOMLY GENERATE
NITAL. POPULATION
AND EVALUATE THE
PERFORMANCE INDEX
. SELECTlON
l CROSS-OVER
. MUTATION
GEN. =G&.+l
I
Fig. 3. Genetic algorithm flow chart
[8]. The equations describing the steady-state operation
of a synchronous generator connected to an infinite bus
through an external reactance can be linearized about
any particular operating point as follows:
APm-AP=iVl~
(3)
AP = K, Ab + K, AE:,
(4)
Fig. 4. The GA computed PSS parameter K for the entire loading
Fig. 6. The GA computed PSS parameter K for the entire loading
range (P, Q)= (0.1. . . 1.0; - 0.3, . . 1.0) (ISE).
range (P .Q) = (0.1, . . 1.0: - 0.3. . . 1.0) (ITAE).
Fig. 5. The GA computed PSS parameter T for the entire loading
range
(P. Q)=
(0.1, .__, 1.0; -0.3, . . 1.0) (ISE).
AE:, =
KS
AEM -
K&
1 + ST&K,
1+ ST&K,
A6
AV,=K,Ad+K,AE;
(6)
The constants K,-K6 are given in Appendix A.
The system data are as follows:
Machine (p.u.):
x,,= 1.6; x; = 0.32;
xy = 1.55
L, = 1 o:
w,,= 1207~ rad ss; TX = 6.0 s
D = 0.0; M= 10.0
Transmission line (p.u.):
r, = 0;
x, = 0.4
Exciter:
K, = 50.0;
T, = 0.05 s
Loading (p.u.)
P = (0.1, 0.2 )..., 1.0);
Q = ( - 0.3, - 0.2 ,..., 1.0)
The supplementary stabilizing signal considered in
this paper is one proportional to electrical power. A
derivative-type power stabilizer is used. Its transfer
function is given by:
G,(s) =
KS
(s + l/7J2
30
1
-
8/9/2019 Tunning Pss Using GA
4/7
140 Y.L. Ahdel-Magid. M.M. Dawoud- Electrw Pmwr System Research 39 (1996) 137- 143
0.8
0.6
+ 0.4
0.2
0
1
Fig. 7. The GA computed PSS parameter T for the entire loading
range (P, Q) = (0.1. .._, 1.0: -0.3. ,,., 1.0) (ITAE ).
where K and Tare the PSS parameters to be optimized.
This particular configuration of the PSS was chosen as
the result of extensive studies of root-locus diagrams
for numerous system loading conditions [lo] .
In this paper, the optimum values of the parameters
K and T, which minimize the different performance
indices selected, are easily and accurately computed
using genetic algorithms. For a given operating point,
the equations describing the machine, excitation con-
trol, transmission line and the PSS, are simulated and
the performance index evaluated. In a typical run of the
GA, an init ial population is randomly generated. This
init ial population is referred to as the 0th generation.
Each individua l in the init ial population has an associ-
ated performance index value. Using the performance
index information, the GA then produces a new popu-
lation. The application of a genetic algorithm involves
repet itively performing two steps:
1. The calculation of the performance index for each of
the individuals in the current population. To do this,
the system must be simulated to obta in the value of
the performance index.
2. The genetic algorithm then produces the next gener-
ation of individuals using the selection, crossover
and mutation operators.
These two steps are repeated from generation to gener-
ation until the population has converged. producing the
optimum parameters. The procedure is then repeated
for the selected set of grid points in the real poweqreac-
tive power domain (P=O.l, 0.2, . . . 1.0; Q= -0.3.
- 0.2, . ., 1.0). A flow-chart of the genetic algorithm
optimization procedure is given in Fig. 3.
Table 1
Variation of T with P and Q (ISE)
P=O.l P = 0.4 P=O.7
P= 1.0
Q = -0.3 1.56
0.38 0.26 0.27
Q = 0.0 1.14 0.61 0.38 0.3
Q = 0.3 0.89 0.7 0.48 0.4
Q = 0.7 0.75
0.68 0.58 0.49
Q= 1.0 0.76 0.7 0.63 0.54
Table 2
Variation of K with P and Q (ISE)
P=O.l P = 0.4
P = 0.7 P= 1.0
Q = -0.3
1.69 I .03 2.26 3.07
Q = 0.0 5.14 1.68
1.93 2.66
Q = 0.3 9.50 2.74 2.3
2.56
Q = 0.7 15.32 4.26 3 2.84
Q= 1.0 18.99 5.19
3.47 3.08
Penalizing only speed excursions, the two perfor-
mance indices considered in this study take the form:
s
-1
ISE = (Au) dt
(8)
0
ITAE =
i
-J tlA& dt
(9)
0
To compute the optimum parameter values, a unit step
disturbance in mechanical power was used to perturb
the system from its operating point.
4. Results and tests
The GA computed PSS parameters are plotted in
Figs. 4-7 in three-dimensional form for the entire
load ing range considered. while Tables l-4 list the
values of these parameters for some selected loading
conditions. Tables 1 and 2 show the optimum values of
K and T when the performance index is the ISE. The
optimum values of K and T for the ISE case are shown
in Figs. 4 and 5 for the entire loading range.
Tables 3 and 4 show the optimum values of K and T
when the performance index is the ITAE. The optimum
values of K and T for the ITAE case are shown in Figs.
6 and 7 for the entire loading range.
It is clear that the optimal settings of the PSS
parameters vary widely with the operating poin t and
also depend on the performance index being minimized.
Simulation tests for various operat ing points and with
the GA computed optimum parameter settings were
performed. Figs. 8- 13 are sample simulation results
illustrating some significant aspects of the system re-
sponse. A step-like disturbance in mechanical power
was used to perturb the system from its operating
Table 3
Variation of ;r with P and Q (ITAE)
Q= -0.3
Q = 0.0
Q = 0.3
Q = 0.7
Q= 1.0
P=O.l P = 0.4
P = 0.7 P= 1.0
0.74 0.21 0.18 0.20
0.58 0.34
0.24 0.23
0.47 0.38 0.29 0.26
0.42 0.38 0.34 0.31
0.41 0.39
0.37 0.36
-
8/9/2019 Tunning Pss Using GA
5/7
Y.L.. Abdel-Magid, M.M. Duwoud Elrcrric Power S~.stem.s Reseurch 39 (1996) 137- 143 141
Table 4
Variation of K with P and Q (ITAE)
P = 0.1 P = 0.4
P=O.7 P= 1.0
Q =
-0.3 2.01
1.58
2.54 3.13
Q = 0.0 5.87 1.85 2.18 2.75
Q =
0.3
10.8
2.97 2.48 2.69
Q=O.7 16.94 4.55 3.19 2.98
Q= 1.0 20.99 5.55 3.68 3.2
point. The traces shown are for speed deviations and
they are for three cases:
1. PSS disconnected.
2. PSS parameters are those resulting from minim izing
the ISE.
3. PSS parameters are those resulting from minim izing
the ITAE.
From the results, it is observed that:
1. The PSS considered significantly increases the stabil-
ity of the system.
2. Altering the PSS parameters to cope with realjreac-
tive power loading maintains optimum performance
for a wide range of system operating conditions.
3. The ITAE is superior to the ISE as far as the
damping of the oscillations and the settling time are
considered.
It is worth noting that the results obtained for the
ISE case agree with those available in the literature [lo],
where Lyapunov functions and more elaborate classical
optimization techniques were used. The results obtained
for the ITAE have not been previously obtained due to
the complexity of problem formulation. Because of the
simplicity of using the GA with complex objective
functions and nonl inear systems, other performance
indices and ful l system representations are being cur-
rently investigated.
: ..
.
-. :
-*:
-0.03
time, s
0
2 4 6 8 10
Fig. 8. Spee d deviation for (P. Q) = (0.1. - 0.3): no stab ilizer (. ):
Fig. 10 . Spee d deviation for (P. Q) = (0.1, 1.0): no stabiliz er (. .);
ISE (p -); ITA E (--).
1SE (m ): ITAE(----).
0.02
A%.of5.4
;T
: . .:
. . >
0.01
0.005
0.
-0.005.
-0.01 -
-0.015
-
: :
;:
:;
i
time,s 1
-0.02L
0 2
4
6 8 10
Fig. 9. Spee d deviation for (P, Q) = (0.4, 0.3): no stabiliz er (. ):
ISE (- I: ITAE (p,.
5. Conclusions
The tuning of power system stabilizers (PSS) using
genetic algorithms (GA) is investigated in this paper. A
digital simulation of a linearized model of a single-ma-
chine infinite bus power system at some operating point
is used in conjunction with the genetic algorithm opti-
mization process.The integral of the square of the error
and the time-multiplied absolute value of the error
performance indices are considered. For each perfor-
mance index, the PSS parameters are optimized off-line
for a selected set of grid points in the real power-reac-
tive power (P-Q) domain. It is suggested hat the PSS
parameters be updated based on measurements of the
generator real and reactive power. Time domain simu-
lations of the system with GA-tuned PSS show the
improved dynamic performance under widely varying
load conditions and points to the superiority of the
ITAE performance index relative to the ISE.
time, s
1
-0.021
0 2
4 6 8 10
-
8/9/2019 Tunning Pss Using GA
6/7
142
Y.L. Abdel-Magid, M.M. Dawoud /Electric Power Systems Research 39 (1996) 137-143
0.06l
tiwfe,s
0 2 4 6 8
10
Fig. 1 I. Speed deviation for (P. Q) = (1.0, -0.3): no stabilizer
(. .); ISE (- - -): ITAE (---).
Acknowledgements
The authors acknowledge the support and encour-
agement of King Fahd University of Petroleum and
Minerals which made it possible to conduct this re-
search.
Appendix A
All variables with subscript 0 are values of the vari-
ables evaluated at their pre-disturbance steady-state
operat ing poin t from known values of VtO, P, and Q, as
given by Eqs. (lo)-( 16). All variables preceded by A are
deviations of these variables from their values at the
steady-state operat ing point. The constants K,-K6 are
given in Eqs. (17)-(22).
POvto
(10)
(11)
vqo = JV - t&
(12)
0.02
A0
: .* . .
-0.015. ;:
time,s
0 2
4 8 8 10
Fig. 12. Spee d deviation for (P, Q) = (0.7, 0.0): no stabiliz er ( ):
ISE (p -); ITA E (---).
Fig. 13. Spee d devia tion for (P, Q) = (1 .O, 1.0): no stabiliz er (. );
ISE (- - m); ITA E (-).
i Q. + xyi&Q. + xyiio
V Y OYO
E,, = uqo + &xy,, = uqo + &xy
E. = J( vdo + x,i,,) + ( vf l - x&J2. = \i( vdo + x,z,,) + ( vf l - x&J2
6, = tan - , = tan -
(%I + &dqo)%I + x&o)
(c,o - x,&Jc,o - x,&J
K, =, =
x,-xl.,-xl.
E,,E, cos 6,,,E, cos 6,
-Y, + x:,Y, + x:,
lNEo sin 6, +NEo sin 6, +
*L + Xl,L + Xl,
K
2
= E, sin S,E, sin S,
x, + x : ,, + x : ,
xi + x,i + x,
KS = ~S = ~
xc/ + xec/ + xe
K4 = ___
x,1 xl/ E sin ho
x,+x;
K5L3 --
&
vcmE, cos 6, - - -
Vyo
x, + .q v,,
x,+x& VT0
&=- L-
El/o
-Yc+ x& v,,
E, sin 6,
(13)
(14)
(15)
(16)
(17)
(18)
(19)
W)
(21)
G-9)
References
[II
VI
VI
[41
[51
D.E. Goldberg, Genetic Algorithms in Search, Optimization and
Machine Learning, Addison-Wes ley, Reading, MA, 1989.
P.J. Fleming and C.M. Fon seca, Genetic a lgorithms in control
systems engineering, Research Rep. .vo, 470, Department of
Automatic Control and Systems Engineering, University of She-
ffield, UK. March 1993.
J.J. Grefenstette, Optimization of control parameters for genetics
algorithms, IEEE Trans. Syst., Man Cybernet., SMC-I6 (1)
(1986) 122-128.
L. Davis, Handbook of Genetic Algorithms, Van Nostrand Rein-
hold, New York, 1991.
P.L. Dandeno, A.N. Karas, K.R. McClymont and W. Watson,
Effect of high-speed rectifier ex citation systems on generator
stability limits, IEEE Trans. Power Appar. Syst., PAS-87 (1968)
190.- 196.
-
8/9/2019 Tunning Pss Using GA
7/7
Y.L. Abdel-Magid, M.M. Dawoud/ Electric P ower Systems Research 39 (1996) 137-143
143
[6] F.P. de Mello, L.N. Hannett and J.M. Undrill, Practical ap-
proach to supplementary stabilizing from accelerating power,
IEEE Trans. Power Appar. Syst., PAS-97 (1978) 1515-1522.
[7] E.V. Larsen and D.A. Swann, Applying power system stabiliz-
ers,
IEEE Trans. Power Appar. Syst., PAS-100 (6)
(1981) 3017-
3046.
[8] F.P. de Mello and C. Concordia, Concepts o f synchronous
machine stability as affected by excitation, IEEE Trans. Power
Appar. Syst., PAS-88
(1969) 316-327.
[9] M.K. El-Sherbini and D.M. Mehta, Dynamic systems stabil-
ity-investigation of the effect of different loadings and excita-
tion systems,
IEEE Trans. Power Appar. Syst., PAS-92 (1973)
1538-1546.
[IO] Y.L . Abdel-Magid and G.W. Swif t, Variable structure power
stabilizer to supplement static-excitation system, Proc. IEE, 123
(7) (1976) 697-701.
[l l] W.C. Schultz and V.C. Rideout, Control system performance
measures: past, present and future, IRE Trans. Automa tic Con-
trol AC-6, 22 (1961) 22-35.
[12] K. Ogata,
Modern Control Engineering.
Prentice-Hall, Engle-
wood Cli ffs , NJ, 1970, pp. 296-313.